In practice
You send the question to 200 customers and 100 answer. 45 answer 9 or 10, 35 answer 7 or 8, and 20 answer 0 to 6. Your NPS is 45% minus 20%, which is 25.
The comments under the score are often more useful than the number. That's where the reason sits.
NPS isn't a satisfaction score. Its useful role is showing how your own number changes over time, and how you compare with direct competitors.
How to measure it
- Promoters: 9 or 10. Passives: 7 or 8. Detractors: 0 to 6.
- NPS = % promoters − % detractors. Passives count in the total but are not subtracted.
- Report it as a whole number, not a percentage.
Common mistakes
- Reading small samples as trends. With 30 answers, two people changing their minds moves the score by about 7 points.
- Comparing your number to another industry's. Bain's own guidance is to compare against direct competitors.
- Collecting the score and skipping the follow-up question, "What's the main reason for your score?"
What changes in AI-first products
A score tells you loyalty moved. It doesn't tell an agent what to change. The "why" comments are the part a coding agent can use, once they're grouped by problem and tied to the customers who wrote them. Annsa takes survey answers in through CSV or its in-product surveys and keeps each comment verbatim next to the priority it supports, so Claude Code, Cursor, Codex or Grok can read the reasons over MCP. A person still decides which reason is worth a build.
See how Annsa does it: Surveys →